Medior/Senior Computer Vision Engineer [startup]

Trosort; Brussel; Hybride; EUR 90 000 - 120 000; Plein temps

Avantages offerts par ce poste

Résumé du poste

Trosort in Brussels, Belgium is seeking a Senior CV Engineer to own real pieces of the stack for on-prem inference. You will ship CV models on-prem, validate performance with real numbers, and bring solutions to production in weeks, while maintaining data quality and robust monitoring.

You’ll work across embeddings, classic CV, transformers, and detection/segmentation, with a push to deploy on real hardware in sorting centers and edge devices. English is required; French or Dutch a plus.

Qualifications

• Minimum 3y of experience in CV
• Shipped CV models on-prem, not just behind a cloud API
• Real depth across model families and the judgment to know which one to reach for
• A builder: can go from idea to tested POC in days, and bring it into production in weeks
• A team player who's also ready to take the lead on a task when it's needed
• Solid grasp of CV model architectures
• Experience validating data and model quality, not just training and hoping
• Experience with TensorRT, Docker, MLOps, Linux, CI/CD

Responsabilités

• Deploy real-time models on-prem, with the constraints that come with real hardware, not just a notebook running on a GPU
• Work across the full toolbox: embeddings, classic CV, transformers, detection/segmentation
• Go from idea to a working POC in days, and validate it with proper A/B testing before it ships
• Keep up with the research: read the papers, test new architectures early
• Own things in the cloud and in production
• Guard the data. Build and trust data validation — garbage in, garbage out, so we don't let garbage in
• Talk before we ask. Forward communication. We hear about problems and progress from you, not from a status meeting

Connaissances

CV models on-prem

TensorRT

Docker

MLOps

Linux

CI/CD

Data validation

Model architectures

Production deployment

Team leadership

Outils

TensorRT

Docker

CI/CD

Linux

MLOps

Description du poste

WAT, Brussels (Belgium) (on-site, max 2d/week remote) | Competitive salary + Stock Options (ESOP) | Python · Computer Vision · TensorRT

Every model you ship helps reuse and keep more garments out of landfill.

What Trosort does

We build the tech stack (hardware + software) that sorts secondhand clothes at scale (defect detection, quality grading, garment recognition, resale value) running on real hardware in real sorting centers. Machines in production, today.

The job

Join our CV Engineering team and own real pieces of the stack. Pick the right model for the problem, ship it on-prem for realtime inference, prove it works with real numbers, keep it working after it ships.

• Deploy real-time models on-prem, with the constraints that come with real hardware, not just a notebook running on a GPU.


• Work across the full toolbox: embeddings, classic CV, transformers, detection/segmentation. Choose what fits the problem.


• Go from idea to a working POC in days, and validate it with proper A/B testing before it ships.


• Keep up with the research: read the papers, test new architectures early.


• Own things in the cloud and in production.


• Guard the data. Build and trust data validation — garbage in, garbage out, so we don't let garbage in.


• Talk before we ask. Forward communication. We hear about problems and progress from you, not from a status meeting.

What we need from you

• Minimum 3y of experience in CV


• Shipped CV models on-prem, not just behind a cloud API


• Real depth across model families and the judgment to know which one to reach for


• A builder: can go from idea to tested POC in days, and bring it into production in weeks


• A team player who's also ready to take the lead on a task when it's needed


• Solid grasp of CV model architectures


• Experience validating data and model quality, not just training and hoping


• Experience with TensorRT, Docker, MLOps, Linux, CI/CD

Nice to have

• Experience in industrial CV


• Experience with industrial cameras


• Background in R&D and object tracking

What you get

An 8-person startup: flat, fast, allergic to bureaucracy, obsessed with making customers happy.

You own your work, no nine layers of approval to fix something.

A genuinely nice place to work (hello WAT).

Practicalities

• Fulltime permanent position


• Diversity is a plus here, come as you are.


• Candidates must already hold EU work authorization


• Language is English (French or Dutch is a plus)


• We work on hardware, no full remote possibilities